AI content and search: the real risk for a small site

The question I get most often about AI and search is whether Google penalizes AI written content. The answer has been consistent for a while now and it is not the one people expect. Google’s stated position is that it judges usefulness, not production method. What it acts against is scaled content abuse, meaning large volumes of pages with little value to anyone.

That distinction matters, because it means the risk is not the tool. The risk is the behavior the tool makes easy.

What actually gets sites in trouble

Scaled content abuse has been the most aggressively enforced spam policy for the last couple of years. In practice it catches a recognizable pattern:

• Dozens or hundreds of pages published in a short period, all covering near identical topics
• Location pages where only the city name changes
• Articles that answer a search query by restating what is already on the first page of results
• Content with no author, no experience, no specifics and nothing that could only have been written by this business
• Pages published unedited, which is usually visible in the first two paragraphs

Google also confirmed in 2026 that its spam policies apply to AI Overviews and AI Mode, not only to the classic list of blue links. So the same behavior that gets you demoted in search also keeps you out of the AI answers, which are an increasing share of how people find things.

Why the incentive changed

The old logic was volume. More pages, more chances to rank, more traffic. That logic worked when the search result was a list and being present on the list was most of the battle.

Now a large part of query volume gets answered in a generated summary that cites a handful of sources. Many sites are reporting lower click through from search alongside growth in referrals from AI tools. I would not put a number on that for your business, because it varies enormously by industry and query type, and you should check your own analytics rather than any published average. But the direction is clear enough to plan around.

What follows from it: being one of many adequate pages on a topic is worth much less than it used to be. Being the source that a system will cite is worth more. Those two things pull in opposite directions from a content volume strategy.

What gets cited instead

What appears to travel well into AI answers is the same set of things that made a page genuinely useful before: information that is specific, verifiable, consistent with other sources, and attributable to a real entity.

• Original information. Your own prices, your own process, your own data, results you actually observed. A model cannot get this anywhere else.
• Clear structure. A direct answer near the top, real headings, short factual statements. Content that is easy to extract is easy to cite.
• Consistent business details everywhere. Name, address, phone, service area, hours, matching across your site, your profiles and directories. Inconsistency makes you an unreliable source to quote.
• Attribution. A named author with a real reason to be credible, a visible date, and a way to check the business exists.
• Genuine reviews and mentions from sources that are not yours. Third party corroboration is what turns a claim into a fact.

How to use AI on content without taking the risk

The safe pattern is straightforward, and it is mostly about where in the process the tool sits.

1. Never let AI decide what to publish. Topics come from real customer questions, sales calls, and search terms in your own account. That is the part that makes content worth having.
2. Supply the substance yourself. Before drafting, write the specifics: the numbers, the process, the objections, the example. If you cannot supply anything a competitor could not, do not publish the page.
3. Use it for structure and first drafts, not for facts. Outlines, transitions, tightening. Every factual claim is checked by a person.
4. Edit properly and cut hard. If a paragraph would be true for any company in your industry, delete it.
5. Publish at a human rate. One good page a week beats forty in a month, both for the search systems and for the reader.
6. Keep your business information identical everywhere. This is dull and it is one of the highest return tasks in the list.
7. Track what happens. Watch impressions and clicks in Search Console, and check whether your brand and your site appear when you ask the main AI tools questions in your category.

The realistic assessment

For most small businesses the danger is not a manual penalty. It is spending months producing pages that never earn anything, then concluding that content does not work. The cost is the wasted quarter, not the punishment.

There is a second cost that is easier to miss. Publishing thirty generic pages puts thirty pieces of average writing under your name. Anyone evaluating you can read them. Content is a sample of your thinking, and a weak sample is worse than none.

What to do next

Look at what you have already published and sort it into two piles: pages that contain something only your business could say, and pages that do not. Improve or remove the second pile. That single exercise usually does more than the next ten articles would.

Then write one page properly on the question your customers actually ask most. If you want help deciding which questions those are and what the page should say, book a free consultation at /contact/.